The Science of Spaced Repetition: Why Our SM-2 Flashcards Actually Work
We implemented the SM-2 algorithm that powers Anki — but with AI-generated cards. Here's the cognitive science behind why it remembers for you.
Memory is a function of how often you retrieve something, not how often you see it. That single insight — from Ebbinghaus in 1885 — is why spaced repetition works.
What SM-2 does
SM-2, the algorithm behind Anki, schedules each flashcard for review right before you'd forget it. You rate the card (Again / Hard / Good / Easy) and the algorithm adjusts two values — an ease factor and an interval — so the next review lands at the edge of your recall.
It's boring math. But the behavioral payoff is that you spend most of your review time on the 10% of cards that are about to slip, not the 90% you already know.
Why AI-generated cards are different
Anki is great but the friction is real: you have to make the cards yourself. For a class 12 biology syllabus, that's thousands of cards.
Coachingle generates them from any topic you type. Key terms, definitions, and mnemonic-style prompts are extracted and formatted as spaced-repetition cards automatically. Quality isn't as hand-crafted as a well-made Anki deck, but the volume-to-effort ratio is dramatic.
What we got wrong
Our first cut ignored the "Again" rating — every failed card restarted from day 1. That tanked long-term retention. Fix shipped in v1.3: failing a card now uses a lapse-specific interval that's shorter but not full-reset, matching SM-2's actual spec.
If you're serious about long-term retention, don't just consume cheatsheets — add the terms to flashcards and review daily for 5-10 minutes.